Sahana-M/self-driving-car

Applied Deep Learning, Computer Vision and Machine Learning techniques to Build an Autonomous Car with Python

This project of Self driving car has been broken down into sections where in each section I have got to learn many things.

In glimpse -

• Used Computer Vision techniques via OpenCV to identify lane lines for a self-driving car.

• Learned to train a Perceptron-based Neural Network to classify between binary classes.

• Learned to train Convolutional Neural Networks to identify between various traffic signs.

• Trained Deep Neural Networks to fit complex datasets.

• Mastered Keras, a power Neural Network library written in Python.

Builded and trained a fully functional self driving car to drive on its own!

Started with these topics ,

---

Computer vision -

• Detecting lanes using OpenCV
• covering algorithms like, Canny algorithm , Hought transform
• link - https://github.com/Sahana-M/self-driving-car/tree/master/finding-lanes

---

Perceptron -

• Learnt how a perceptron works and how to build a model using Keras
• Which factors decide the efficiency of a model
• learnt sigmoid activation function , Adam optimizer, Binary cross entropy for binary classification
• Solved a simple prediciting a diabetic patient or not use case using this model
• Used only a single hidden layer
• link - https://github.com/Sahana-M/Keras-implementations-DNN-CNN/blob/master/perceptron23.ipynb

---

Keras classification -

• This time working on a difficult data set which is a non linear data set
• Used Keras to classify this dataset
• This time with more layers of DNN , 2 Hidden layers
• link - https://github.com/Sahana-M/Keras-implementations-DNN-CNN/blob/master/deep_neural_network.ipynb

---

Multiclass classification using Keras -

• Implemented multiclass classification using Keras
• This time used One hot encoding, softmax activation function , and categorical cross entropy as it is multiclass classification
• link - https://github.com/Sahana-M/Keras-implementations-DNN-CNN/blob/master/multiclass_classification.ipynb

---

DNN MNIST Image Recognition -

• Used Deep Neural networks on MNIST data set to recognise Handwritten digits using Keras
• Used relu and softmax activation functions
• Model obtained an accuracy of ~93
• link - https://github.com/Sahana-M/Keras-implementations-DNN-CNN/blob/master/MNIST%20Deep%20learning.ipynb

---

CNN MNIST Image Recognition -

• Used Convolutional neural networks this time on MNIST data set to recognise Handwritten digits
• This time did a bit of image preprocessing, One hot encoding
• Used LenNet based model to classify our MNIST data set
• accuracy ~99%, better than DNN
• link - https://github.com/Sahana-M/Handwriting-recognition-using-Convolutional-neural-network-using-Keras

---

Classifying Road signs -

• dataset - https://bitbucket.org/jadslim/german-traffic-signs
• Used modified LeNet based model
• Used various measures to improve the accuracy -
  • fine tuning
  • dropout layers
  • fit generator
  • data augmentation
• link - https://github.com/Sahana-M/self-driving-car/tree/master/identifying%20traffic%20signs

---

Polynomial Regression -

• A simple implementation of polynomial regression using Keras which is very much necessary for Self driving car concepts
• link - https://github.com/Sahana-M/self-driving-car/blob/master/Polynomial_regression.ipynb

---

Behavioural Cloning -

• The most important part of self driving car
• Did required image preprocessing
• data augmentation using Batch generator
• Fine tuning the model
• Used NVIDIA Neural model
• depolyed the model n Udacity self driving car simulator
• link - https://github.com/Sahana-M/self-driving-car/tree/master/Behavioural%20cloning%20-%20Nvidia%20neural%20model